Nvidia GT200b rumours and speculation thread

http://www.vr-zone.com/articles/Nvidia_Prepares_For_40nm_GPUs/6020.html

Could this info mean that NVIDIA has canceled GT200 in 55nm and go straight to 40nm a few months later? Maybe that`s why there is no new info about it`s 55nm GT2xx parts.

40nm is still at least 6 months away.
Doing a -relatively- simple 55nm optical shrink of the GT200 core now would be cheaper for them, and it would still not prevent their 40nm parts from doing business as scheduled later on.
Why cut costs only 6 months from now, if you can start it right away (albeit in a diminished form) ? They did it with the G9x parts.
 
According to Neuschäfer they won't rename the card because the modifications are not enough to justify a new name.
I assume this is the same logic that switched GT and GTS in the mobile space?
 
So GT200b really is dead? Because I thought the launch date was August 25?

Anyway depending price, I would be really interested in a 216 SP GT260. As of now 260 trades blows with 4870 pretty evenly, but it does have a lot more VRAM and typically can be found a bit cheaper on newegg (frequently as low as $230 after MIR, versus 250 for the cheapest 4870), giving it the possible edge for me. A 216 SP for the same price would cinch it, at that point it would probably have a clear 5-10% performance edge, 896 MB RAM, run cooler etc.

But if they just end up pricing it in tweener land at $280 or something, then it's just a normal old boring case of getting what you pay for, and not compelling.

I have to think if the 216 SP GT260 is out at the same price as current 260 (big if) it would force AMD to drop the 4870 price as well. Something they should have a lot of flexibility to do.
 
Well, it says GeForce GTX 200:

nvidia_gpu_2013_3.jpg
 
Heh, that's just a GTX 280 with the S1070's shader clock speed. It could mean more than that, but call me skeptical...
 
Isn't the Tesla C1060 around that ALU performance?

Jawed

That's what Nvidia was touting at Editor's Day, yes. They'd fullfill their promise of a TFLOP-card (after a fashion) not necessarily with a consumer line product.
 
How will they achieve +1 TB/s in 2013? That would require 16 GHz Memory at 512 bit or 8 GHz at 1024 bit.
I'm sure they don't know themselves, but if you extrapolate from the past, it's not entirely unreasonable.

Google once highlighted a technology speculation thread of the early nineties on usenet where forum particants tried to predict future. It turned out that people generally underestimated the effective numbers. (Especially wrt clock speeds and storage capacity.)

It doesn't matter anyway if they're off by a factor of 2 or 3, that's not really the point.
 
So could it mean NVIDIA has problems with it`s GT200@55nm part?

During NVidia's quarter results conference call they indicated that 55nm is already in production, so it's not "problems" in that sense. It's simply that they also indicated that they have a lot of 65nm left to run through as well. One can tell that the 4000-series of cards has thrown them off on a number of fronts, from ASP targets/predictions to demand/supply issues as well.
 
Back
Top